The GenAI ecosystem is fragmented. Thousands of tools, frameworks, and platforms — each solving a piece of the puzzle, none connecting them all. Until now.

The Problem​

Every AI team faces the same challenges:

• Tool sprawl — MCP servers, LangChain, Semantic Kernel, AutoGen, CrewAI — each with its own patterns
• No standard wiring — How do you compose an agent with a skill, constrain it with instructions, and guard it with hooks?
• Reinventing the wheel — Every new AI project starts from scratch instead of building on proven architectures
• Quality gaps — No systematic way to ensure AI solutions meet reliability, security, and cost targets

The Solution: FrootAI​

FrootAI is the uniFAIng glue — an open-source platform that connects the dots:

🧩 AI Primitives​

Six composable building blocks:

🔗 FAI Protocol​

The open specification (fai-manifest.json) that wires primitives together. Like package.json for npm, but for AI architectures:

{
  "play": "01-enterprise-rag",
  "version": "1.0.0",
  "context": {
    "knowledge": ["./docs/rag-patterns.md"],
    "waf": ["reliability", "security", "cost-optimization"]
  },
  "primitives": {
    "agents": ["./agents/rag-architect.agent.md"],
    "skills": ["./skills/build-rag-pipeline/SKILL.md"],
    "instructions": ["./instructions/azure-search.instructions.md"]
  }
}

🏗️ Solution Plays​

Pre-built, production-ready AI architectures. Each play includes:

• DevKit — Copilot-ready agents, skills, instructions
• TuneKit — Customer-tunable AI parameters
• SpecKit — Documentation and metadata
• Infra — Azure Verified Modules (Bicep)

📦 Distribution Everywhere​

Install FrootAI wherever you work:

# MCP Server
npx frootai-mcp@latest

# Python
pip install frootai

# VS Code
ext install frootai.frootai

# Docker
docker pull ghcr.io/frootai/frootai-mcp

What Makes FrootAI Different?​

1. Protocol-level composition — Primitives auto-wire when used inside solution plays
2. Well-Architected by default — Every primitive aligns to 6 WAF pillars
3. Framework-agnostic — Works with LangChain, Semantic Kernel, AutoGen, and more
4. Production-ready — Not demos, but deployable architectures with infrastructure

Get Started​

npx frootai-mcp@latest

Then explore the documentation, browse solution plays, or dive into the FROOT learning path.

What's Next​

We're building in the open. Star us on GitHub, try a solution play, or contribute a primitive.

The uniFAIng glue for the GenAI ecosystem. 🍊